What is a deep net

We said we will talk extensively about layers.

Time to keep our promise Here’s arguably the most common pictorial representation of deep neural networks

this is our first layer.

It is called the input layer.

That’s basically the data we have.

We take the inputs and get outputs as we did before.

The main rationale behind neural networks however is that we can now use these outputs as inputs for

another layer and then another one and another until we decide to stop the Last Lear we build is the

output layer.

That’s basically what we compare the targets to.

All right.

So the first layer is the input layer and the last layer is the output layer.

All the layers between are called hidden layers we call them hidden.

As we know the inputs and we get the outputs but we don’t know what happens between as these operations

are hidden stacking layers one after the other produces a deep network or as we will call it a deep

net.

The building blocks of the hidden layer are called hidden units or nodes.

Here’s a hidden unit in mathematical terms if h is the tense or related to the hidden layer each hidden

unit is an element of that tensor.

The number of hidden units in a hidden layer is often referred to as the width of the layer usually

but not always we stack layers with the same with so that the we’re with is equal to the width of the

entire network OK.

We saw how wide a deep network is.

Let’s examine how deep it can be.

Depth is an important ingredient as it refers to the number of hidden layers in a network.

When we create a machine learning algorithm we choose it’s width and depth we refer to these values

as hyper parameters hyper parameters should not be mistaken with parameters.

Recall that parameters were the weights and the bias’s hyper parameters are the with depth learning

rate and some of the variables we will see later.

The main difference between the two is that the value of the parameters will be derived through optimization

while hyper parameters are set by us before we start optimizing.

All right.

This will do for now.

Thanks for watching.